Which customers are less or more loyal to an international bank?

An Exploratory Data Analysis on Jupyter Notebook

Nowadays, customer churn prediction in retail banking is getting attention since every customer plays an important role in the competition among other banks. Customer retention is crucial as acquiring new customers is often more costly than keeping the current ones. Therefore, customer churn analysis is vital to the success of a company and the first step to understanding the customer profile. Reducing the number of customer churn is possible with the help of clear understanding of the profile of the churners.

With this purpose, I applied Exploratory Data Analysis (EDA) on an international bank churn data set which was obtained from Kaggle.

You can see the Jupyter notebook file and the presentation of the study.

GitHub: https://github.com/ezgicn/Project_EDA

Kaggle: https://www.kaggle.com/ezgicn/bank-customer-churn-eda/notebook

Methodology

1. Specify potential factors which affect on churn decision
2. Determine potential customers who tend to churn
3. Identify loyalty degree of the customers

What are the potential churn factors?

• Country

Country distribution on churners

It is seen that customer churn rates in Germany and France are 2 times higher than in Spain.

• Gender

Gender distribution on churners

As can be seen on the pie graph, the female churner rate is higher than the male churner rate.

• Number of product

Country and number of product distribution on churners

The heat map indicates that the churned customer number decreases in all countries when the number of products increases.

• Credit Card

Credit card effect on churn decision

Interestingly, churn decision among customers is higher even they have a credit card.

• Being an active member

Active member effect on churn decision

Churner number is lower among customers who are active members.

The average age of churned customers who are not active members and don’t have a credit card is higher. The age range of customers having 1 product is very large. There is no non-churner even they have 4 products if they don’ have a credit card and are not active members.

The average age of churned customers who are active members and have a credit card is highest if they have 4 products.

• Age Group

Balance & Age groups effect on churn decision

Customers, who are under age 25 & age 65 and older in Germany and France, tend to churn more with respect to balance and credit score.

Credit score & Age group effect on churn decision

In addition, credit score, tenure, balance, and estimated salary don’t affect churn decision. Still, I would like to point on the balances in Germany where there is no zero balance in their account.

Conclusion

Less Loyal Customers :(

• in Germany & France
• Female
• Number of product < 2
• Age 25–65
• Has credit card
• Student & retired customers in Germany & France

More Loyal Customers :)

• in Spain
• Male
• Number of product ≥ 2
• Active member
• Student & retired customers in Spain

Future Work

• Negative, zero, and positive balance effect would be investigated if there was negative balance data of the customers.
• The relation between churn decision and product types would be analyzed if there was information about data types in addition to the number of products.
• Machine Learning algorithms can be applied to predict churn decision of the customers.